Skip to content

CLAUDE.md

Latest commit

 

History

History
74 lines (51 loc) · 3.41 KB

CLAUDE.md

File metadata and controls

74 lines (51 loc) · 3.41 KB
Raw

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Commands

# Install dependencies
npm install

# Run in production
npm start

# Run in development (auto-restarts on file changes via nodemon)
npm run dev

There are no tests configured. There is no lint script.

Environment Setup

Copy .env.example to .env. The required vars are:

  • LLM_PROVIDER — one of openai, anthropic, gemini, local
  • LLM_MODEL — model identifier (e.g. claude-sonnet-4-6)
  • The API key matching the chosen provider (ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, or LOCAL_MODEL_BASE_URL)
  • OPENMRS_BASE_URL, OPENMRS_USERNAME, OPENMRS_PASSWORD

The server exits at startup with a clear error if LLM_PROVIDER or LLM_MODEL is missing or the provider is not recognised.

Architecture

An Express proxy server that sits between an OpenMRS frontend and LLM provider APIs. Its sole purpose is to keep API keys server-side rather than exposing them to the browser.

The server uses tool-calling (function calling): the LLM is given a set of clinical tools and decides which ones to call to fetch the data it needs from OpenMRS, then produces a structured visit summary.

Entry point: server.js — app setup, CORS, routing, and startup. Provider logic lives in providers/.

server.js                      ← app setup, CORS, routing, startup
providers/
  utils.js                     ← shared throwIfNotOk helper and DEFAULT_TEMPERATURE
  openai.js                    ← OpenAI /v1/chat/completions (tool-calling)
  anthropic.js                 ← Anthropic /v1/messages (tool-calling)
  gemini.js                    ← Google Gemini generateContent API (tool-calling)
  local.js                     ← any OpenAI-compatible local server (tool-calling)
clinical-tools/
  definitions.js               ← CLINICAL_TOOLS array + toOpenAITools / toGeminiTools converters
  executor.js                  ← ToolExecutor — dispatches tool calls to visit-data fetchers
  visit-data.js                ← OpenMRS REST/FHIR fetch helpers for each tool
prompt-template.js             ← buildSystemPrompt helper

Single endpoint: POST /api/generate-visit-summary

  • Request body: { visitUuid: string, patientUuid: string }
  • Response: { summary: string }
  • Provider and model come from LLM_PROVIDER / LLM_MODEL env vars, resolved once at startup
  • visitUuid and patientUuid are baked into the ToolExecutor context; the LLM calls tools to fetch data on demand

Adding a new provider: create providers/<name>.js exporting an async callWithTools(model, systemPrompt, tools, executor) function, then add it to the PROVIDERS_WITH_TOOLS map in server.js.

CORS is enforced via the ALLOWED_ORIGINS env var (comma-separated). Requests with no Origin header (curl, server-to-server) are always allowed.

Health check: GET /health returns { status: 'ok' }.

All upstream API calls use Node's native fetch (no SDK).

  • OpenAI: /v1/chat/completions
  • Anthropic: /v1/messages
  • Gemini: https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent
  • Local: ${LOCAL_MODEL_BASE_URL}/chat/completions (OpenAI-compatible format)

OpenMRS credentials (OPENMRS_BASE_URL, OPENMRS_USERNAME, OPENMRS_PASSWORD) are read from env vars and used by clinical-tools/visit-data.js to call the OpenMRS REST/FHIR APIs.